US 7158591 B2
A filter and processing sequence is described that efficiently combines and performs two or more tasks required to demodulate a composite 3G (third generation) wireless signal formed by a combination of wideband 3.84 MHz (Universal Mobile Telecommunications System, identiied a acronym “UMTS”, or Universal Mobile Telecommunications System Terrestrial Radio Access, identified as acronym “UTRA”) carriers and narrowband 1.2288 MHz CDMA-2000 carriers. The three tasks, applied to each spectral component of the 3G wireless signal and described in the order of a traditional filtering structure are: Spectral translation, Bandwidth Reduction, and Sample Rate Selection. These tasks are traditionally implemented in three distinct pieces of hardware or software modules.
1. A receiver for receiving and efficiently separating a composite 3-G wireless communications signal into constituent baseband components, wherein said receiver combines multiple processing tasks of a 3-G receiver into a single filter, the single filter comprising,
an equal ripple linear phase recursive filter channelizer performing simultaneous spectral translation and bandwidth reduction of multiple channels,
an equal-ripple linear phase recursive filter interpolator for performing sample rate changes,
whereby the equal ripple linear phase recursive filter channelizer performs the processing tasks required to operate multiple channels, including spectral translation, bandwith reduction, sample rate changes, and outputs a separate time series for each channel in the composite 3-G wireless communications signal.
A filter and processing sequence is described that efficiently combines and performs two or more tasks required to demodulate a composite 3G (third generation) wireless signal formed by a combination of wideband 3.84 MHz (Universal Mobile Telecommunications System, hereinafter referred to as “UMTS” or Universal Mobile Telecommunications System Terrestrial Radio Access, hereinafter referred to as “UTRA”) carriers and narrowband 1.2288 MHz CDMA-2000 carriers. The three tasks, applied to each spectral component of the 3G wireless signal and described in the order of a traditional filtering structure are: Spectral translation, Bandwidth Reduction, and Sample Rate Selection. These tasks are traditionally implemented in three distinct pieces of hardware or software modules.
The spectrum processed by the receiver is shown in
When describing the processing technique presented herein we will use sample rates selected to satisfy the timing consideration. The sample rate indicated in
The tasks and the associated modules that implement the functions in a traditional receiver are: (1) Spectral Translation, (2) Bandwidth Reduction, and (3) Output Sample Rate Selection. The Spectral Translation is performed by a complex heterodyne that translates the center of the desired spectral band to base-band. The complex heterodyne multiplies the input data sequence by samples of a cosine wave and a sine wave with frequency selected to match the center of the desired band. The Bandwidth Reduction is performed by a digital filter that processes the complex input data stream of the down-converted signal. The digital filter performs the required weighted sums to form the reduced bandwidth output data stream. The digital filter performs a low-pass filtering process that restricts the signal bandwidth to that of the translated band, and consequently rejects the remaining spectral components of the translated signal. The Output Sample Rate Selection is performed by a complex digital filter known as an interpolator that accept input data from the previously described low-pass filter at a fixed input rate that satisfies the Nyquist Criterion, and computes from these samples a set of output samples at an output rate different from the input rate and selected to satisfy some signal conditioning constraint in subsequent processing following this processing block.
The invention described here combines two or more of the processing tasks described above in a single filter, and further has the single filter perform the tasks for more than one center frequency signal. The filter structure is the well-known polyphase partition. In this structure a single filter is partitioned into M-parallel paths each representing a section of the prototype filter. The outputs of these paths are combined with fixed phase rotators to obtain separate time series from the multiple center frequency bands of interest. In this parallel path structure, different center frequencies only affect the set of scalar phase rotators associated with each path. Thus the combination of polyphase partition with their post filter phase rotators permits the single filter to operate at each of the filter bank center frequencies simultaneously.
In an equivalent, but alternate structure, a complex band pass filter replaces the low pass filter. This filter has an impulse response formed as the product of the low pass impulse response h(n) and the up-converting complex heterodyne sequence exp(jωoTsn) of the same length. Here the heterodyne is applied to the filter to move its center frequency to the band center of the signal rather than the standard approach, which applies the heterodyne to the signal to move its band to the filter's spectral location. In this structure, the filtering occurs at the signal's center frequency, and the output of the filter is properly band limited but still resides at the carrier center frequency. If it is desired to translate the signal spectra to base band, this down conversion can be applied after the filter as shown in
Since the signal bandwidth has been reduced by the band limiting action of the digital filter, it is common to reduce the sample rate of the down converted and filtered time series. The heterodyne following the band pass filter can be moved to the low data rate side of the down sampler. Now only the samples delivered to the output of the down sampler are subjected to the heterodyne and the workload of the heterodyne is reduced by the same M-to-1 ratio of the input to output sampling rates. The down sampling operation is thus applied to the band-centered signal. Reducing the sample rate of the carrier centered signal results in an alias induced spectral translation of the center frequency from fc with angular rotation rate of 2π fc/fs per sample to an angular rotation rate of 2π M fc/fs modulo(2π). If the center frequency fc is any multiple of the output sample rate, say k fs/M, then the aliased rotation rate is 2π M (k fs/M)/fs modulo(2π) or k 2π modulo(2π) which is congruent to zero, which means the output rate of rotation is zero radians per sample. For the proper choice of center frequency relative to sample rate, the down sampled data samples represent a signal that has been aliased to DC. Selection of the sample rate to be an integer multiple of the signal center frequency is one of the suggested restrictions addressed earlier. The restriction is also applicable when the ratio of sample rate to center frequency is a rational ratio of small integers.
Any multiple of the output sample rate will alias to DC or zero frequency. Similarly, any offset from a multiple of the output sample rate will alias to the same offset from DC or zero frequency. A heterodyne following the down sampling can then remove this residual offset. Thus the spectral translation from the channel center can be accomplished at the filter output prior to the down sampling or after the down sampling by a combination of aliasing and reduced data rate heterodyne. The sliding of the output heterodyne to the downside of the output resampler is shown in
When a large number of phase rotators are required to service multiple channels, they are implemented as a fast Fourier transform (FFT). We still have the option to perform the output heterodyne prior to down sampling, or after, and we still have the option to perform the down sampling, with an input commutator, prior to the filter segments rather than after. System considerations related to the interpolation requirements following the filtering and down conversion influence where the down sampling operation is to be performed. Computational efficiency is increased as the down sampler is moved towards the input data stream. Moving the down sampler to the input of the process results in the structure shown in
One final consideration in this class of polyphase filter partitions is that the down sampling that occurs via the input commutator can be modified to permit M-to-P sample rate change as opposed to the traditional M-to-1 change. This is accomplished by replacing the weight vector for each polyphase stage with a set of weight vectors that are cyclically accessed with period P while the stages are accessed with period M. This permits the sample rate change, normally allocated to a second interpolation filter, to also be imbedded in the polyphase filter. Thus the polyphase filter can, with proper attention to resampling, partitioning, and weight-set scheduling, accommodate the translation, filtering, and sample rate change of the entire filter bank process described in
Because of the two classes of signals to be processed, the polyphase filter structure can be implemented as a cascade of processing tasks in two modes. In the first mode, the filter is implemented as two independent parallel structures with one performing the processing required to implement or service the needs of all three wideband signals and one performing the processing required to implement or service the needs of the eleven narrowband signals. In the second mode, the filters operate in cascade with one filter partitioning the full input bandwidth into overlapping spectral bands matched to the spectral width of the three wideband components and one filter processing the reduced bandwidth signals obtained from the first filter to obtain, if required, additional bandwidth reduction matched to the narrowband signals. These configurations are shown in
Following the philosophy that a filter should operate at the lowest possible sample rate consistent with its Nyquist rate, we will emphasize the second option composed of cascade processing tasks interspersed with appropriate resamplers between bandwidth reducing stages.
A set of digital filters composed of polyphase partitions of prototype low pass filters coupled with a process for performing sample rate changes within the filtering process is applied to the task of performing the simultaneous functions of channelizing, of filtering, and of resampling a frequency division multiplexed communication signal. In particular, the signal is a third generation (3G) signal suite composed of mixes of wideband UTRA (3.84 MHz) and narrowband (1.2244 MHz) spectral components with bandwidths and center frequencies shown in
The collection of Multirate signal processing partitions and scheduling presented here take advantage of signal bandwidths and signal center frequency locations and separations to enable a single processing function to simultaneously perform bandwidth control, spectral translation, and resampling for separate channels with similar and related spectral characteristics. The process afford great reduction in processing load required to demodulate the multiple channels comprising the 3-G signal set.
The channelization system is first described at a high level by a collection of interconnected functional processing blocks assigned to perform specific processing tasks. The signal processing performed by a particular block may represent the entire processing required to extract a desired signal component from the composite signal, or it may represent one of a sequence of signal processing functions required to extract the desired signal. In general, the processing is performed in a hierarchical cascade of high level processing blocks. These blocks can be described by interconnections of lower level processing blocks that are common to many of the high level blocks. The filtering blocks are traditionally non-recursive because of the general ease with which the prototype low pass filter can be decomposed into polyphase segments. A particular class of recursive filters that permit the polyphase partition of its prototype low pass realization can also be used to form the processing blocks.
The recursive structures often exhibit spectral responses that occasionally require post processing spectral clean-up filters. These clean-up filters are unique to the recursive implementations, and represent additional processing blocks not present in the non-recursive implementation. The incentive to use a recursive implementation for the filtering blocks is the significant reduction in processing required for a given filtering task. The recursive polyphase filter can be implemented with structures that offer linear phase response, a property required to preserve signal fidelity. The recursive polyphase filter can also be implemented with non-uniform phase, with a marked reduction in processing workload. This option is viable when the receiver includes a channel equalizer that will attribute the filter phase distortion to the channel and correct it while inverting the channel response. The non-linear phase recursive polyphase structure offers additional flexibility in parameter selection for the cascade processing tasks. Implementations that mix and match from the three realization options can also offer design flexibility. This patent only describes the non-recursive implementation. Related and connected patents describe the recursive only, and the mixed non-recursive and recursive implementations.
First Processing Block:
The first processing block of this invention is shown in
Three fundamental processing blocks shown in
The six-channel channelizer is composed of a pair of 3-channel channelizers, The upper 3-channel channelizer extracts the three wideband UTRA channels centered at −5 MHz, 0 MHz, and +5 MHz. The lower 3-channel channelizer extracts the overlapped bands centered at −2.5 MHz and a +2.5 MHz. This lower segment of the processing block is disabled when the signal set is composed of the three-wideband UTRA carriers. As stated, the initial processing block of the resampling 6-channel Channelizer is a pair of three-stage linear-phase recursive polyphase filters (P-610) that is shown in
The three-stage polyphase filter (P-610), shown in
The three output samples from the output port of the 3-stage polyphase filter are presented to a 3-point DFT (P-620). This is a standard numerical algorithm that can be implemented directly as a collection of Inner Products, as a factored FFT, or as a reduced multiplication Winograd Transform.
The five retained output samples from the pair of 3-point DFT are time samples representing translated, filtered, and re-sampled signals from each of the center frequency bands described earlier. These include the three-wideband signals comprising the 3-G signal set as well as the overlapped bands centered at ±2.5 MHz that are only used in the last two frequency assignments shown in
The polyphase channelizer described just described has been operated with an output rate equal to its input rate of fs (15.36 MHz) The next processing step in the receiver chain is the resampling of the signals to obtain an output sample rate of ⅖ fs, (6.144 MHz). This step is accomplished by the 1-to-2 interpolating filters (P-640) followed by a 5-to-1 down sampler. This structure is shown in
Since the half-band filter is designed as a polynomial in Z2, and knowing that two units of delay at the output rate is the same as one unit of delay at the input rate, we can interchange the order of 1-to-2 up-sampling and filtering. This is shown in
When the two-path half-band, linear phase, and now re-sampled structure is inserted in
The interaction of the two commutators can be described in a simple scheduling routine that is listed in table 1. From this table we see that the five input samples are partitioned into two subsets, one of three successive samples, and one of two successive samples. For each input sample we must exercise the recursive filter stage, and after three input samples we take an output sample from the lower path, and after the next two input samples, we take an output sample from the upper path. Note that 5 inputs requires 10 products for a workload of two products per input point, and when we extract our two outputs at the cost of ten products, we determine the output workload at five products per output sample.
Note that a single stage recursive prototype filter constrained by architecture to have linear phase has been used to filter and translate all three wideband channels as well as, via the second channelizer, the overlapped narrowband signals bands in preparation for further processing. The filter is a three-path filter requiring 9 products for each of two paths, and zero operations for the top path. Let us compare the savings attributable to the polyphase filter partition. In the direct implementation, a 90-tap prototype low pass FIR filter will meet the transition bandwidth and out-of-band attenuation requirements for the broadband channel decomposition. Similarly, a 10-tap prototype low pass FIR filter will meet the transition bandwidth and out-of-band attenuation requirements of the up-2, down-5 interpolator. Using these lengths as a benchmark, we can compare the relative workloads, in equivalent complex operations, of the two options, direct processing and the technique described here. A complex operation (comp-op) is considered to be a complex scalar multiply and add, requiring two real multiplies and adds. In the direct implementation, the workload to process 5 input samples is 20 comp-ops for the input heterodynes, 1350 comp-ops for the three filters, and 60 comp-ops for the interpolators. The workload for all three channels is 1430 comp-ops per 5-inputs or equivalently 1430 comp-ops per 2-outputs. Normalizing by the number of channels and number of data points we obtain a workload of approximately 95 comp-ops per input data point per channel or 238 comp-ops per output data point per channel. In the polyphase resampling implementation the workload to process 5 input samples is 90 comp-ops for the filter, 20 comp-ops for the DFT, 20 comp-ops for the output heterodynes, and 30 comp-ops for the interpolators for a total of 160 comp-ops per 5-inputs or 160 comp-ops per 2-outputs. Normalizing by number of channels and number of data points we obtain a workload of approximately 11 comp-ops per input data point per channel or 27 comp-ops per output data point per channel. The relative workload for the direct versus the polyphase resampler is approximately 9-to-1.
Second Processing Block
We now consider the second processing block of this invention, the resampling 5-channel channelizer that is shown in
The 3-channel resampling channelizer uses the 6-channel resampling channelizer as a preprocessor and initial bandwidth and sample rate reducer. This arrangement is shown in
The resampling 5-channel channelizer can be based on a non-recursive filter structure or on a recursive filter structure. A related patent (Resampling Digital FIR Filter Structure for Demodulating 3G Wireless Signals) addresses the non-recursive option. In this embodiment, we address the recursive option.
Three fundamental processing blocks of the resampling 5-channel channelizer are shown in
The initial processing block of the narrowband 3-channel Channelizer is the complex heterodyne (P-820) that moves each of the three (or one) narrowband CDMA-2000 signals to base band. The base band signal is then presented to the 5-to-2 interpolator (P-840) that reduces the bandwidth to match the desired output sample rate and then performs the interpolation to that rate by a combination of up-sampling and down-sampling. The interpolator, presented with moderate detail in
The spectrumin the first figure, counting left-to-right, is the prototype, 4-multiply half band filter formed as polynomials in Z2. The spectrum shown in the second figure is the zero packed, hence replicated spectra, obtained by operating a copy of the first filter but operating with 4-delays (Z4) per stage rather than 2-delays (Z2). This filter is also a 4-multiply (zero-packed) half band filter. The spectrum shown in the third figure is that obtained by cascading the two previous filters. The pass band of the composite corresponds to the intersection of their separate pass bands, hence is defined by the zero-packed filter. This composite filter exhibits linear phase over its bandwidth and has a transition bandwidth and out-of-band attenuation level that, if implemented as a FIR filter, would require 108 taps. The filter as shown requires 8-multiplies per input data sample. The 4-th and last figure presents the log-magnitude and phase response of the total composite filter, including the 1-to-2 up sampling and the 5-to-1 down sampling operation. The entire composite filter has a workload, exclusive of the down sampling, of 10-operations per input sample or 5-operations per output sample. Since the down sampler discards 4-out-of-5 samples, the workload per retained sample is 25 operations per output sample. Other combinations of down sampling, polyphase half-band filtering, and up sampling can be assembled to effect the same bandwidth reduction and sample rate change. Advantages of this class of two-path, linear-phase, recursive, half-band polyphase filters over traditional FIR and IIR filters are:
The embodiment of the invention currently preferred by the inventor has been described, but one skilled in the art of digital signal processing and digital receiver design will be enabled by acquaintance with the foregoing disclosure to design a number of alternative embodiments of this invention, and this should be borne in mind when construing the scope of the claims which follows this specification.